Posted on 02/03/2023 6:01:20 PM PST by SeekAndFind
OpenAI, the maker of Artificial Intelligence chatbot ChatGPT, announced on Tuesday that it has released a new software tool to help detect whether someone is trying to pass off AI-generated text as something that was written by a person.
The tool, known as a classifier, comes two months after the release of ChatGPT, a chatbot that generates human-like responses based on the input it is given. Schools were quick to limit ChatGPT’s use over concerns that it could fuel academic dishonesty and hinder learning, as students have been using the chatbot to create content that they are passing off as their own.
OpenAI researchers said that while it was “impossible to reliably detect all AI-written text,” good classifiers could pick up signs that text was written by AI. They said the tool could be useful in cases where AI was used for “academic dishonesty” and when AI chatbots were positioned as humans.
In a press release, OpenAI warns the classified’s public beta mode is “not fully reliable,” saying that it aims to collect feedback and share improved methods in the future.
The firm admitted the classifier only correctly identified 26 percent of AI-written English texts. It also incorrectly labeled human-written text as AI-written 9 percent of the time.
The classifier also has several limitations, including its unreliability on text below 1,000 characters, as well as misidentifying some human-written text as AI-written. It also only works in English for now, as it performs “significantly worse in other languages and it is unreliable on code.” Finally, AI-written text can be edited to evade the classifier, according to OpenAI.
“It should not be used as a primary decision-making tool, but instead as a complement to other methods of determining the source of a piece of text,” OpenAI said.
ChatGPT is a free program that generates text in response to a prompt, including articles, essays, jokes, and even poetry.
Since ChatGPT debuted in November 2022 and gained wide popularity among millions of users, some of the largest U.S. school districts have banned the AI chatbot over concerns that students will use the text generator to cheat or plagiarize.
Following the wave of attention, last week Microsoft announced a multibillion-dollar investment in OpenAI, a research-oriented San Francisco startup, and said it would incorporate the startup’s AI models into its products for consumers and businesses.
Reuters contributed to this report.
Exactly how many lines of code is this ChatGPT?
“Take Note College Students!”
~~~~~~~
Take note #EneMedia!
RE: Exactly how many lines of code is this ChatGPT?
From Quora:
https://www.quora.com/What-s-the-code-for-Chat-GPT-look-like-How-many-lines
The code for GPT (Generative Pre-trained Transformer) and its variants such as ChatGPT are complex and consist of many lines of code. The codebase for GPT-3, for example, is estimated to be around 175,000 lines of code.
The code for GPT-3 is written in Python, and it uses the PyTorch library for its machine learning and deep learning operations. The codebase consists of several different components, including the preprocessing and tokenization of input text, the training of the model, and the generation of text.
The main component of the GPT-3 model is the transformer architecture, which uses self-attention mechanisms to process input sequences. The transformer is implemented using PyTorch’s nn. Module and nn.Transformer classes, which are built on top of PyTorch’s tensor operations.
It also uses a number of other libraries and modules, such as Hugging Face’s tokenizers library for tokenization, and OpenAI’s parameter server for distributed training.
It’s open-source, so you can check the code in GitHub.
It’s worth noting that GPT models are trained on a massive amount of data and require a significant amount of computational resources, and running the model at scale requires specialized hardware and infrastructure.
"ChatGPT, write my paper so that the OpenAI thinks it is written by a human."
I imagine bad spelling would be a giveaway.
Given the way kids write these days, it’d be easy to get around this plagiarism check:
- Add “This means” to every other sentence
- Change the font from paragraph to paragraph
- Randomly delete commas
- Randomly add commas
- Randomly add semicolons
Too easy.
“It also incorrectly labeled human-written text as AI-written 9 percent of the time.”
So only 1 out of 10 innocent students get kicked out of school.
Use synonyms and re-arrange sentences.
There will be false positives.
Good teachers will know if something is up.
At one of the sites we support, I was talking with a Professor that has had a small but nagging problem over the years of students copying care plans. They have started to select students suspected of such cheating to “defend” their care plans in person. She said that almost all who cheat in this manner falter when they have a minimum of notes to refer to at such a defense.
The Prof said faculty in other disciplines are moving to oral Q/A on submitted papers.
Re: 9 - It usually goes way beyond that. I’ll suggest that students that use such tools don’t know the material or subject. An oral defense of their work will show that pretty quickly.
Run it against Reuters. I’ll bet more than half of their news articles are machine-generated.
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.